MiniMaxMiniMax Music

MiniMax Music 3

Music 3 generates music from a description and lyrics. Its published architecture separates global musical structure from local acoustic detail, then synthesizes stereo audio.

Several open components contribute to the system

The licence identifies Qwen3-8B as the language-model ancestor. It also identifies Stable Audio code as the starting point for DiT-2B and Descript's audio codec code as a basis for the VAE.MiniMax Music3 Community License

Local output differs from hosted settings

The card documents several local deployment routes. Its Diffusers guidance describes full precision below 24 GB of VRAM, with CPU offloading paths for smaller cards. Those are configuration-specific claims, not a guarantee for every runtime.MiniMax Music3 model card

Judge the complete musical phrase

Practical guidance. Listen across section changes, repeated lyrics and sustained notes. A strong first few seconds can hide unstable structure later in the track. Keep lyrics separate from the style description so that each can be revised independently.

Specifications & access

MiniMax Music 3 specifications
DetailValue and conditions
AccessDownloadable, API, Provider product
Model identifiersmusic-3.0MiniMax-Music3music-3.0-free
AncestryThe global language model starts from Qwen3-8B. The synthesis path combines global and local hidden states.MiniMax Music3 model card
Local outputThe documented local API returns 32 kHz, 16-bit stereo WAV audio. Its text prompt is limited to 5,000 tokens.MiniMax Music3 model card
Hosted accessSince 20 August 2026, paid music APIs are limited to existing paying users. The free API tier was discontinued; weights and MiniMax Audio remain alternatives.MiniMax music access notice and workflows
Commercial weight termsVisible MiniMax-Music3 attribution is required. Aggregate yearly revenue above US$20 million from relevant products or services, including affiliates, requires written authorization.MiniMax Music3 Community License

Related models

  • MiniMax Music 2.6

    MiniMax · MiniMax Music

    A previous-generation text-to-music model with supplied lyrics, automatic lyrics and instrumental generation.

    Launch date not confirmedRestricted accessAPI
  • A music model that generates a new rendition from reference audio and a target style.

    Launch date not confirmedRestricted accessAPI